Converting an approved purchase request into an official order is one of the tasks where automation has already made significant inroads. Not because AI understands something special here, but because the task itself leaves little to understand. The request has already been approved, the supplier is fixed, the format of the order is fixed. What remains is: taking over data, generating a document, sending it, recording it in the ERP. That is the kind of work for which rules suffice and for which no judgment is needed about what should actually be ordered.
Three axes determine the picture here. The structuredness is high: the input is unambiguous, the output is a fixed document, no interpretation is needed of vague wording or incomplete information. The volume is high: companies of some size place dozens to hundreds of orders per week, often following the same steps. And the room for judgment is limited, because the decision to order has already been made by the person who approved the request. What remains is execution, not decision-making.
Against this, customer contact and physical actions play hardly any role, which makes things easier: there is no one who needs to be convinced of the order, and there is nothing that needs to be physically moved to complete the task. Creativity is virtually absent, which makes sense for a task that is by definition repeatable. Error costs and compliance are in the middle: an incorrectly ordered quantity or a wrong supplier number costs money and time to correct, and in regulated procurement (think of hazardous substances or certified parts) additional requirements apply. That is the reason why human oversight in practice usually continues to exist, even when execution is automated.
The technology that handles this task is not generative AI in the sense of writing text or creating images, but RPA: robotic process automation. A system that reads the approved request, transfers the data into an order format, creates the document and sends it to the supplier, and writes the status back into the ERP. That works well as long as two conditions apply: there is a fixed order format, and the approval flow is linked to the system that generates the order. Without that link, a human remains the bridge between approval and execution, and most of the freed-up hour disappears again.
This is therefore not a task where AI itself decides what is ordered. That decision — how much, from whom, under what conditions — lies with the purchase request that has already been approved. The task that is taken over here is the administrative step after that: from request to order form, without a human having to type that in manually.
At a company with a single ERP system, a limited number of fixed suppliers and a standardized order format, this task can largely be automated today already. At a company that orders through separate emails, by phone, or with varying formats, that is different: standardization must happen first before automation yields anything. The nature of what is purchased also plays a role. Standard office supplies or regular raw materials are easy to automate; custom orders with specifications negotiated per order more often require a human keeping an eye on things.
This task is also not separate from the rest of the procurement process. What happens before the order is placed — how inventory levels are monitored to determine whether reordering is needed — partly determines how many orders arise and how predictable that volume is. And what happens after placement, such as drawing up shipping documents for the delivery or processing the purchase invoices that follow the order, belongs to the same chain of administrative steps that can be automated well once the systems connect to each other.
This is not an assessment of what a company should do with freed-up hours, and not an argument for or against a staffing decision. If automating this task has consequences for positions or staffing levels, the applicable statutory requirements regarding labor law and works council involvement apply; that is a different question from whether the task itself is technically capable of being taken over.
Whether this task has the described profile in your own organization depends on your ERP setup, the number of suppliers, and the extent to which order formats are already fixed. A broad initial indication is provided by the free quickscan: twelve questions, without an account, with an indication of what portion of the hours in this kind of work can be taken over by AI today. The full work scan, which breaks down the work of an entire company into tasks and calculates per task the FTE capacity, is still under construction.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.